Reinstatement of long-term memory following erasure of its behavioral and synaptic expression in Aplysia

Thus, chelerythrine treatment reversed LTS, and TSA treatment blocked chelerythrine's reversal of LTS. There were no significant differences among 5XTrained-Veh, 5XTrained-TSA and 5XTrained-TSA-Chel groups, nor between 5XTrained-Chel and Control-Veh groups, at 48 hr. Asterisks, comparisons of the Control-Veh group with the 5XTrained-Veh group, the 5XTrained-TSA group, the 5XTrained-Chel group, and the 5XTrained-TSA-Chel group at 24 hr; and of the Control-Veh group with the 5XTrained-Veh group, the 5XTrained-TSA group, and the 5XTrained-TSA-Chel group at 48 hr. Plus signs, comparisons of the 5XTrained-Chel group with the 5XTrained-Veh group, the 5XTrained-TSA group, and the 5XTrained-TSA-Chel group at 48 hr. (C) Experimental protocols. The times of the pretests, training, posttests, and drug/vehicle injections are shown relative to the end of the last training session. The intrahemocoel injection of either TSA or vehicle is indicated by the red arrow. Note that TSA was injected into animals prior to the sensitization training. (D) Inhibiting histone deacetylation facilitated the induction of LTS in Aplysia. The mean duration of the SWR in the 3XTrained-TSA group (n = 4) at 24 hr was 59.8 ± 0.3 s, whereas it was only 2.0 ± 1.0 s in the 3XTrained-Veh group (n = 4; p < 0.001, unpaired t test). Asterisks, comparison between 3XTrained-TSA and 3XTrained-Veh groups; Error bars represent ±SEM.DOI: http://dx.doi.org/10.7554/eLife.03896.014

Previous work has shown that LTF of in vitro Aplysia sensorimotor connections depends on alterations of chromatin structure. In particular, 1X5HT, which normally induces only short-term facilitation, was found to induce LTF when applied with TSA (Guan et al., 2002). In intact Aplysia three bouts of tail shocks are insufficient by themselves to induce LTS (Control-Veh-3XTrained data, Figures 7 and 8). We tested whether three bouts of shocks could induce LTS if delivered in the presence of an HDAC inhibitor. Animals were given three bouts of shocks 15 min after an intrahemocoel injection of either vehicle solution or TSA (Figure 9C). 24 hr later the animals that received three bouts of shocks after an injection of TSA exhibited LTS, whereas animals that received the shocks after an injection of vehicle did not (Figure 9D). Thus, HDAC inhibition facilitates the induction of LTM in Aplysia.

The present morphological results, together with those of our previous behavioral and electrophysiological investigations (Cai et al., 2011, 2012), suggest that the persistence of sensitization-related LTM in Aplysia does not require the persistence of the synaptic connections generated during learning. Rather, LTM appears to be regulated by a homeostatic mechanism that specifies the net synaptic strength according to experience. Following 5X5HT stimulation, we observed that the number of presynaptic varicosities in the sensorimotor cocultures increased to a new overall value, and that this value was maintained despite the appearance and disappearance of individual varicosities. Furthermore, blockade of memory reconsolidation, or inhibition of PKM, reset the number of varicosities to the original, non-facilitated value. Remarkably, this resetting did not result from the straightforward retraction of the varicosities that appeared by 24 hr after 5X5HT stimulation, varicosities whose growth coincides with LTF (Glanzman et al., 1990); rather, the resetting was produced by a complex reorganization that involved retraction of both 5HT-induced and original varicosities, as well as growth of new, additional varicosities. Thus, for any given sensorimotor connection, the homeostatic regulatory mechanism appeared indifferent to the identities of the particular presynaptic varicosities and, presumably, active zones (Schacher et al., 1990), that mediated baseline synaptic transmission and synaptic facilitation. Our data argue that synapses are not ‘tagged’ with respect to memory storage, at least for a connection involving a single SN and a single MN; the data may therefore reflect a fundamental distinction between the cellular and molecular processes of LTM storage and those of LTM induction in Aplysia, in which synaptic tagging plays a critical role (Martin et al., 1997).

The data also suggest that synaptic change is an expression mechanism, rather than a storage mechanism, for LTM in Aplysia (although see below).

Importantly, neither reconsolidation blockade nor inhibition of PKM caused the varicosity number to drop below the starting value. (Notice that the same is true for the amplitude of the EPSP in our earlier synaptic physiological studies, which used the same training regimens [Cai et al., 2011, 2012].) This result provides additional evidence for the operation of a homeostatic regulatory mechanism, and suggests that the homeostatic mechanism toggles between two all-or-none states, a facilitated, sensitization-related state and a nonfacilitated one. (There must also be a third state, synaptic depression, in which the strength of the sensorimotor connection is reduced below the nonfacilitated level [Montarolo et al., 1988].) Our results are reminiscent of those from studies in the Drosophila neuromuscular junction and the mammalian CNS, which also indicate that synaptic plasticity is regulated by homeostatic mechanisms (Turrigiano, 2007).

The present conclusions appear to conflict with those from morphological investigations of LTM in mice in which dendrites have been chronically imaged in the intact, living brain. Several such studies have reported stability of some dendritic spines for periods of weeks-to-months after training (Xu et al., 2009; Yang et al., 2009; Liston et al., 2013). These results, as well as evidence that spines in the adult mammalian cortex exhibit little turnover (Grutzendler et al., 2002), have led some to argue that stable spines provide a physical basis for the lifelong maintenance of memories (Bhatt et al., 2009). But other studies of the mouse brain using very similar in vivo imaging techniques have reported a high degree of spine instability in the adult cerebral cortex (Trachtenberg et al., 2002; Holtmaat et al., 2005); the contradiction between these two sets of opposed findings remains unresolved. Furthermore, even in those studies where spines of adult cortical neurons were reported to be highly stable, there was significant instability of both pre-existing and new spines within the first 2–4 days after a training protocol (Xu et al., 2009; Yang et al., 2009), and we monitored SN varicosities for only 48 hr after training with 5HT. Therefore, the present results are not necessarily inconsistent with those from in vivo imaging studies in the mouse brain.

We found that LTS could be fully reinstated following its disruption by the same antimnemonic treatments previously shown to eliminate LTF (Cai et al., 2011, 2012; Lee et al., 2012; Hu and Schacher, 2014) (Figures 7 and 8). This finding is consistent with two possible explanations. First, the LTM for sensitization may be intact in animals following reconsolidation blockade and inhibition of PKM, despite the elimination of both behavioral enhancement and synaptic growth. According to this scheme, synapses serve merely to express LTM, they are not sites of LTM storage. The reinstatement of LTM expression, then, involves restoring the appropriate number of synapses between the SNs and MNs, as determined by an experience-registering homeostatic mechanism. The second possible explanation incorporates the notion of an as-yet-unidentified priming mechanism; here, reconsolidation blockade and inhibition of PKM, can erase the stored sensitization memory through the reversal of pre- and postsynaptic structures induced by the long-term training, but the antimnemonic treatments do not eliminate the primer. The primer does not constitute LTM, but is required for its reconstitution via new synaptic growth. The priming signal might interact with fresh facilitatory input to the withdrawal circuit, due to the additional (3×) tail shocks, to upregulate the number of synaptic contacts; the appropriate number of contacts could be determined by the homeostatic process, and involve signals from existing and new synapses.

In a previous study we attempted to reinstate LTS following its disruption by inhibition of PKM Apl III but failed (Cai et al., 2011) (see Figure 3B of the earlier study). Interestingly, we used only a single bout of tail shocks in that attempt. In light of the present results (Figures 7 and 8), our previous failure indicates that one bout of tails shocks is insufficient to fully recruit the signaling pathways that reinstate LTS (Abel et al., 1998; Sossin, 2008; Mayford et al., 2012; Zhang et al., 2012). In future work we will seek to discover how the molecular signals evoked by three bouts of tail shocks differ from those evoked by a single bout.

The data from our behavioral experiments involving the use of an HDAC inhibitor (Figure 9) implicate epigenetic modifications (Levenson and Sweatt 2005; Rahn et al., 2013; Graff and Tsai 2013a)—either in the SN or MN or, most likely, both—in the storage mechanism for LTM in Aplysia. Data from an in vitro study of LTF have also implicated histone acetylation in LTM in Aplysia (Guan et al., 2002). How might HDAC inhibition block chelerythrine's apparent disruption of LTM? Possibly, TSA, by inhibiting histone deacetylase, activates the transcription of genes that promote LTM; the activation of these genes, in turn, may compensate for a disruptive action of chelerythrine on LTM. Findings by others are consistent with this idea. Thus, treatment with an HDAC inhibitor rescues cognitive deficits and learning-related synaptic plasticity in mice that are heterozygous for a mutant form of CREB binding protein (CBP) (Alarcon et al., 2004). In addition, it has been found that phosphorylation of CBP by the atypical PKCζ is required for the histone acetylation that promotes the differentiation of developing neural precursors into neurons, astrocytes, and oligodendrocytes; mutant mice haploinsufficient for CBP exhibit abnormal development of the cerebral cortex (Wang et al., 2010). Furthermore, chelerythrine, by inhibiting the phosphorylation of CBP by PKCζ, blocks the differentiation of cultured precursor cells into astrocytes, and oligodendrocytes (Wang et al., 2010). A recent study in Aplysia has found that reducing the activity of CBP in SNs through intracellular injection of CBP small interfering RNA (siRNA) impaired the induction of LTF in sensorimotor cocultures (Liu et al., 2013).

Interestingly, this deficit in LTF induction could be rescued through the use of a 5HT training protocol designed to maximize the phosphorylation and synthesis of CCAAT/enhancer-binding protein (C/EBP). Taken together, these studies suggest that chelerythrine can disrupt epigenetic processes, possibly involving histone acetylation stimulated by CBP, that promote the induction and maintenance of LTM. Moreover, because HDAC inhibition can block chelerythrine's deleterious effects on memory, as well as promote the induction of LTM, it is intriguing to speculate that these effects might be due to enhancement of a CBP-dependent pathway.

But the above explanation for why TSA blocks chelerythrine's effect on consolidated LTM raises a critical question. Our data suggest that consolidated LTM persists after chelerythrine treatment (as well as reconsolidation blockade), possibly as nuclear changes. If so, then the molecular basis for the persistence of LTM is unlikely to be histone acetylation, because an apparent action of chelerythrine is to reverse histone acetylation. Another prominent epigenetic mechanism known to play a role in LTM is DNA methylation (Day and Sweatt, 2010). Although early evidence indicated that learning-induced DNA methylation in the hippocampus was transient and readily reversible (Miller and Sweatt, 2007), a more recent study has reported that contextual fear conditioning in rats induces DNA methylation of the gene for calcineurin in cortical neurons that persists for at least a month (Miller et al., 2010). Furthermore, Sweatt and colleagues have shown in a rat model of childhood maltreatment that early trauma can produce changes in the DNA methylation of the gene for brain-derived neurotrophic factor (BDNF) in the cortex, changes that persist into adulthood (Roth et al., 2009). Thus, DNA methylation may constitute an epigenetic mechanism for the lifelong storage of memory (Day and Sweatt, 2010). Interestingly, 5HT has recently been reported to induce DNA methylation of the promoter of the transcriptional repressor of memory CREB2, thereby facilitating the induction of LTF (Rajasethupathy et al., 2012). Finally, another potential candidate for a nuclear mechanism of LTM storage is suggested by a recent study by Suberbielle et al. (2013) that found, remarkably, that learning causes DNA double-strand breaks (DSBs) in neurons in the brains of control mice; thus, chromatin remodeling subsequent to DNA DSBs may also encode LTM.

Could a nonsynaptic storage mechanism based on nuclear changes mediate the maintenance of associative memories, particularly those induced in complex neural circuits in the mammalian brain, where a given neuron may have 1000s or 10s of 1000s of synaptic partners? An obvious difficulty confronting any hypothetical nuclear storage mechanism in the mammalian brain is how the appropriate number of connections can be maintained in a synapse-specific manner after learning has occurred. For example, if a synaptic contact that has undergone Hebbian long-term potentiation (Bliss and Collingridge, 1993) as a consequence of associative learning retracts, how could a nuclear storage mechanism restrict the growth of a replacement contact to the correct pair of pre- and postsynaptic neurons? Possibly, there are nonsynaptic ways for neurons to communicate that ensure specificity of associative synaptic plasticity in the face of the significant lability of synaptic structure documented here.

There are, of course, somal, non-nuclear, mechanisms of memory storage or priming that could account for the apparent persistence of memory following synaptic erasure. One such mechanism would be the persistent activity of a kinase (Hegde et al., 1993; Bougie et al., 2012), or persistent inhibition of a phosphatase (Sharma et al., 2003), in the cell body of the SN or MN or in both cell bodies. Another somal mechanism consistent with our data is ubiquitination of one or more somal proteins (Hegde et al., 1997; Chain et al., 1999); up-regulation of a ubiquitin-proteasome pathway could degrade proteins that inhibit the storage of LTM, or that block the priming of LTM.

The present data necessitate a reappraisal of the mnemonic consequences of blockade of memory reconsolidation and inhibition of PKM. It has been previously argued that these manipulations can erase consolidated memory (Nader and Hardt, 2009; Sacktor, 2011; Glanzman, 2013) (although see Lattal and Abel, 2004). But our results indicate that the effect of reconsolidation blockade and PKM inhibition is not to delete LTM but, rather, to impair its expression. In other words, the antimnemonic actions of these manipulations may result from their ability to at least partially reverse the cellular and molecular changes, including synaptic growth, that mediate the expression, rather than the storage, of LTM. Possibly, neither reconsolidation blockade nor PKM inhibition can reverse the nuclear remodeling—involving epigenetic modifications, particularly DNA methylation (Day and Sweatt, 2010; Graff et al., 2011; Rahn et al., 2013), as well as, perhaps, chromatin remodeling subsequent to DNA DSBs (Suberbielle et al., 2013)—that represents the stored memory trace. Alternatively, these antimnemonic manipulations might indeed disrupt the stored memory trace, but leave intact some memory-priming signal. Likely candidates for a residual priming signal include the persistent repression of transcription of the CREB repressor, CREB2 (Bartsch et al., 1995; Upadhya et al., 2004; Liu et al., 2011), and persistent degradation of the regulatory subunit of protein kinase A (PKA) (Chain et al., 1999).

Because chelerythrine's disruption of LTM in Aplysia is blocked by TSA, chelerythrine's antimnemonic actions must involve histone deacetylation. Previous speculation regarding how chelerythrine and the zeta inhibitory peptide (ZIP) disrupt memory maintenance in both Aplysia and rats has focused on the ability of these pharmacological agents to inhibit the phosphorylation of proteins by atypical PKM (Cai et al., 2011; Sacktor, 2011) (Box 1). The present results imply that chelerythrine and ZIP may interfere with epigenetic processes induced by atypical PKM, as well as inhibit protein phosphorylation by atypical PKM. A prior study reported that, following inhibition of PKMζ activity in the insular cortex of rats by the peptide inhibitor ZIP, conditioned taste aversion could not be reinstated by an application of the unconditioned stimulus (UCS, intraperitoneal injection of lithium chloride) (Shema et al., 2007). Our experience (Cai et al., 2011, and the present results) suggests that the failure to reinstate conditioned taste aversion in this study was not because LTM had been extinguished, but because a single application of the UCS was too weak to reverse the disruptive effects of ZIP on mechanisms of LTM expression, some of which may involve chromatin remodeling (Figure 7A,B).

Although these results raise questions about the necessity of PKMζ for maintaining long-term memory in the mammalian brain, as well as the specificity of the peptide inhibitor (ZIP) commonly used to block the activity of PKMζ in mammalian studies (Ling et al., 2002; Pastalkova et al., 2006; Shema et al., 2007), we do not believe they impact the present conclusions. Biochemical tests performed in Aplysia have established that chelerythrine, at the concentration used in our experiments, is a selective and effective inhibitor of PKM (Villareal et al., 2009). Furthermore, 5HT has been shown to activate PKM Apl III via calpain dependent cleavage of the atypical Aplysia PKC, PKC Apl III (Bougie et al., 2012). A possible explanation for the results of the PKMζ knockout mice (Lee et al., 2013; Volk et al., 2013), which lack the gene for the atypical PKCζ (note that in the mammalian CNS PKMζ is not formed from proteolytic cleavage of the atypical PKC, but, rather, is transcribed from an alternate promoter within the PKCζ gene [Hernandez et al., 2003]) is that, in the absence of PKMζ, a second atypical mammalian PKC isoform, PKCι/λ—or its PKM fragment, PKMι/λ—can assume the mnemonic functions of PKMζ (Ren et al., 2013). But there is only one atypical PKC isoform in Aplysia; we therefore believe that chelerythrine's inhibitory effect in the present experiments was selective for PKM Apl III.

In summary, we have found that LTM, or a primer for LTM, can persist following reconsolidation blockade and inhibition of PKM. Furthermore, the residual memory/primer must be independent of synaptic plasticity, because it persists following elimination of the synaptic changes induced during learning. Our results indicate that consolidated memories may be far more refractory to modification or elimination than generally supposed. If confirmed and extended to mammals, the present results would have important implications for treating disorders of LTM, such as posttraumatic stress disorder (PTSD).

The sensorimotor cocultures consisted of one pleural SN and one small siphon (LFS-type) MN. Adult abdominal ganglia and pleural ganglia were removed from 60–100 g Aplysia and then incubated in protease (10 mg/ml Dispase II [Roche Applied Science, Indianapolis, IN] in Leibowitz-15 [L-15, Sigma, St Louis, MO]) for 2 hr at 35°C before desheathing. The appropriate amounts of salts were added to the L15 to yield the following concentrations in mM: 400 NaCl, 11 CaCl2, 10 KCl, 27 MgSO4, 27 MgCl2, 2 NaHCO3. Additionally, the L15 was supplemented with penicillin (50 unit/ml), streptomycin (50 µg/ml), dextrose (6 mg/ml) and glutamine (0.1 mg/ml). After desheathing, SNs and MNs were individually dissociated from ganglia and paired in cell culture. The culture medium contained 50% Aplysia hemolymph and 50% L-15. The cultures were maintained at 18°C in an incubator for 3 day before the start of the experiments. The SNs and MNs were labeled with the intracellular dyes dextran fluorescein and dextran rhodamine B (Molecular Probes, Eugene, OR), respectively, on Day 3/4 in culture. The dyes were dissolved in 0.2 M KCl with 0.25% fast green (final concentration of 10 mg/ml) and then microinjected into cells via brief pressure pulses (6–12 MΩ resistance electrodes, 10–20 psi for 2–5 pulses [40 ms]).

The fluorescent images of the labeled cocultures were acquired with a LSM Pascal (Zeiss, Thornwood, NY) confocal microscope during three imaging sessions: immediately prior to, and 24 hr and 48 hr after, 5X5HT training (or at the equivalent times for Control cocultures). The images were taken using a 20×, 0.5 NA objective. The total number of varicosities was counted for each sensory neuron from the confocal images. All image analyses were performed using Axiovision 4.8.2 (Zeiss, Thornwood, NY). The counter was blind to the experimental conditions of the imaged cocultures. SN varicosities that were in clear contact with, or overlapping, postsynaptic structures (either the MN soma, major neurite, or fine processes) were counted. The majority of the SN varicosities contacted either the MN soma or its initial segment. Only those varicosities having a punctate shape—those in which an oval-shaped body could be distinguished by the narrowing of the neurite on either side—and a measured area of 10 µm2 or greater were counted. If a large fluorescent varicosity comprised several visible punctate varicosities in the image, those small varicosities were counted individually; if not, the structure was treated as a single varicosity. Three categories of varicosities are tracked and quantified: original varicosities, that is the varicosities present at 0 hr; varicosities formed during 0–24 hr; and varicosities formed during 24–48 hr.

Immediately after the first imaging session, some of the cocultures were given 5HT training. 5HT was prepared fresh daily as a 10 mM stock solution in artificial sea water (ASW) and then diluted to the final concentration of 100 µM in the perfusion medium immediately before the first application. The perfusion solution contained 50% L15 and 50% ASW. The 5HT training consisted of five 5 min pulses of 5HT. After each 5 min pulse, the 5HT was rapidly washed out with normal perfusion medium for 15 min. The Control cocultures were treated with the perfusion solution alone. Following 5HT or control treatment, the perfusion medium was replaced with culture medium and the cocultures were returned to the 18°C incubator. A stock solution of anisomycin (Sigma, St Louis, MO) was made by dissolving the protein synthesis inhibitor in dimethyl sulfoxide (DMSO) to a concentration of 40 mM for use in the reconsolidation experiments. After the second imaging session (24 hr after 5HT training or the equivalent time in Control experiments) the anisomycin stock solution was diluted to a concentration of 10 µM in perfusion solution and applied to cocultures for 2 hr. Immediately prior to the anisomycin treatment, some cocultures were given one 5-min pulse of 5HT (100 µM) as the reminder stimulus. The 5HT-containing perfusion medium was washed out with normal perfusion medium in the coclutures that received the reminder. To inhibit PKM Apl III chelerythrine (EMD Biosciences, San Diego, CA) was dissolved in dH2O to a concentration of 10 mM to make a stock solution. After the second imaging session 24 hr after 5HT training, the chelerythrine stock solution was diluted to a concentration of 10 µM in perfusion solution and applied to cocultures for 1 hr.

Following the anisomycin or chelerythrine treatment, the drug was rapidly washed out of the coculture dishes with normal perfusion medium; afterwards the perfusion medium was replaced with culture medium and the cocultures were returned to the incubator.

Adult A. californica (80–120 g) were obtained from a local supplier (Alacrity Marine Biological, Redondo Beach, CA, USA). Animals were housed in a 50 gal aquarium filled with cooled (12–14°C), aerated seawater (Catalina Water Company, Long Beach, CA, USA). The behavioral training and testing methods were similar to those previously described (Fulton et al., 2008; Cai et al., 2011, 2012). Three pretests were performed at once per 10 min, beginning 25 min before the start of training. Full sensitization training consisted of five bouts of electrical shocks delivered to the tail at 20-min intervals. During each bout, the animal received three trains of shocks spaced 2 s apart. Each train was 1 s in duration; the shocks (10-ms pulse duration, 40 Hz, 120 V) were delivered via a Grass stimulator (S88, Astro-Med, West Warwick, RI) connected to platinum wires implanted in the tail. After training the animals were given posttests as indicated in the figures.

A stock solution of 40 mM anisomycin or 10 mM chelerythrine was prepared as for the cell imaging experiments (above). Anisomycin was then diluted in ASW to a concentration of 20 mM (50% DMSO). 200 μl per 100 g of body weight of anisomycin was injected into the animals. Injections of the same amount of vehicle solution (DMSO in ASW) were made in Control experiments. The final concentrations of anisomycin in the animal were approximately 40 μM. The final concentration of DMSO in the hemocoel was ∼0.1%. A volume of 200 μl per 100 g of body weight of chelerythrine was injected into the animals. Trichostatin A (TSA) (Sigma, St Louis, MO) was dissolved in DMSO to a concentration of 10 mM to make a stock solution. To inhibit the histone deacetylase, a volume of 100 μl per 100 g of body weight of TSA was injected into the animals. The specific times at which the intrahemocoel injections were made are indicated in the relevant figures.

All statistical tests were performed using SPSS 22.0 (IBM, Armonk, NY). Parametric tests were used for all statistical analyses. For each coculture the number of varicosities counted at 24 hr and 48 hr after 5HT training were normalized as described in the figure legends. The normalized data were expressed as means ±SEM. For the analysis of the data involving monitoring of synapses/behavior over time, the overall data were first assessed with repeated-measures two-way analyses of variance (ANOVA). If the repeated-measures ANOVA indicated a significant interaction, one-way ANOVAs were performed on the separate test times, followed by Student-Newman-Keuls (SNK) posthoc tests for pairwise comparisons. An unpaired Student's t-test was used to determine the statistical significance of the differences when there were only two groups in the data set (data in Figures 2,5,6 and 9D). All reported levels of significance represent two-tailed values unless otherwise indicated.